[PEAK] Re: setuptools: Test works only after second invocation

[Nikolaus Rath]

"P.J. Eby" <pje at telecommunity.com> writes:
> At 03:38 PM 4/29/2010 -0400, Nikolaus Rath wrote:
>>On 04/29/2010 02:36 PM, P.J. Eby wrote:
>> > At 12:34 PM 4/29/2010 -0400, Nikolaus Rath wrote:
>> >> Is setuptools by any chance doing more of this kind of black magic?
>> >
>> > Sure, lots.  ;-)
>> >
>> >> I'm about to start debugging a second, even more mysterious looking
>> >> problem when running the test command. If it is used in combination
>> >> with Psyco, I get exceptions after all the tests have completed:
>
> Well, it's still a shutdown-time issue - presumably go_lock is being
> set to None before this point.  However, if the only part of
> setuptools you're using is the test command, and it's not installing
> anything from source (and thereby using the sandbox), it's unlikely to
> be a direct result of setuptools.

Actually it does seem to be a direct result of setuptools. When I add
the psyco import to setup.py, the problem goes away. I suppose it's the
same bug that causes the problems with the multiprocessing module.


Is there some more in-depth documentation about what setuptools is doing
than the one on http://peak.telecommunity.com/DevCenter/setuptools?
Something that describes e.g. when, how and why the sandbox is set up
and what is being done to sys.modules? I'm quite curious, and maybe
it'll also help me to track down similar bugs faster or recognize
constructions in my tests that could cause other problems.



Best,

   -Nikolaus

-- 
 »Time flies like an arrow, fruit flies like a Banana.«

  PGP fingerprint: 5B93 61F8 4EA2 E279 ABF6  02CF A9AD B7F8 AE4E 425C